Which projects to vote for in the Pentathlon Race?

[petrusbroder]

OK, folks it is that time of year again.
Before I register the TeAm in the race we need to decide which projects to vote for.
Here is the link to the rules and thus how to select projects.
Discussion is encouraged.

Deadline for registration and voting is April 27, 2018.
Deadline for registration without the power to vote is may 2, 2018.

Thus we have 20 days to discuss and decide ...

Edit:
I have added a poll. The poll is limited to a maximum of 10 choices (it should have been 14 ...). Since Rosetta and Citizen Science grid got such a positive response Those two projects will the TeAm vote for. I've not included Universe, since it did not fit in the poll and since it does not seem stable enough. If that is not OK with you, please let me know in this thread and I'll try to adapt ...

 

[TennesseeTony]

Rosetta, CSG, Universe, and......Asteroids. Or SETI. If we HAVE to do math, Rakesearch.

NFS is pretty demanding on hardware, needs 1GB of RAM per task. YAFU doesn't share resources. PrimeGrid gets enough attention.

 

[Pokey]

As usual I will put my GPUs on whatever comes our way. The only favorite I have is Seti.

I am GPU heavy and CPU light so I will do what I can. Just don't expect any help on the CPU side.

 

[iwajabitw]

GPU for Asteroids is Nvidia only right? If so Ill have to go with PG & Seti for GPU. For CPU as much as I like Universe, it has to be monitored for the hanging tasks that runs on forever. So I like CSG, Rosetta, Asteroids(CPU), and Universe maybe Numbers or NFS if we think Uni will be problematic.

 

[iwajabitw]

Just saw the disclaimer for Astroids GPU

**: Asteroids@home is listed as a CPU project because, in contrast to other GPU projects, a GPU is barely more powerful than a CPU

 

[StefanR5R]

Universe has a new application - old bugs are probably gone, new bugs are in. ;-) I ran Universe only briefly right after their restart, and at that time their result files were so big that my personal bottleneck wasn't CPUs but upload bandwidth, by a factor of 5! Haven't checked back whether they improved this in the meantime.

 

[iwajabitw]

Universe has a new application - old bugs are probably gone, new bugs are in. ;-) I ran Universe only briefly right after their restart, and at that time their result files were so big that my personal bottleneck wasn't CPUs but upload bandwidth, by a factor of 5! Haven't checked back whether they improved this in the meantime.

Good info! I may try some tasks after the Sprint tomorrow.

 

[Ken g6]

I think it might be informative to see our rank in each project in Formula BOINC:

4: Citizen Science Grid
6: Amicable Numbers
8: ODLK
11: Yafu
13: Seti@home
14: Universe@Home
16: NFS@home
16: RakeSearch
17: PrimeGrid
19: Rosetta@home
22: SRBase
24: NumberFields@home
25: Asteroids@home

Therefore, the obvious choices would be CSG, Amicable, ODLK, and YAFU.

The obvious problem with this plan is that YAFU doesn't play well with other projects, or return results quickly. So let's leave that one out.

Then we have an interesting choice. The rules say only one GPU project will be chosen; they don't say we have to vote for only one GPU project. Therefore, we can:

- Vote for one GPU project, and hope all our choices win. It looks like Amicable Numbers would be good for us.
- Vote for two GPU projects to increase our chances of getting one of them. However, the two worse GPU projects have about the same rank for us.
- Vote for no GPU project, let the chips fall where they may there, and vote for four CPU projects. But Amicable Numbers would be much better for us than the other two.

So I think we should stick with CSG, Amicable, and ODLK. If Universe works well, let's vote for that. If Universe doesn't work well, I don't like NFS either. So then would it make sense to vote for SETI as our #2 GPU project, or RakeSearch?

I clearly need to try both Universe and Amicable after the current sprint.

 

[StefanR5R]

Formula Boinc ranks aren't a very good criterion, IMO.

• In the marathons, we are good in projects which big teams ignore throughout most of the year. They won't ignore single disciplines during Pentathlon...

• In the sprints, we compete only with the few teams who actually take part in FB sprints. And the FB sprint crews of the bigger teams are probably smaller than the troops that they are going to bring to Pentathlon.

 

[Ken g6]

Alright, let's look at this another way, what projects do we not want?

In GPU projects, they all look good, SETI, PrimeGrid, or Amicable. I have a personal bias toward one. But overall that probably doesn't matter.

In CPU projects, we probably don't want:
11: Yafu - Slow and unpredictable, uses CreditScrew badly.
14: Universe@Home - Unfamiliar new app uses too much bandwidth.
16: NFS@home - Too much RAM.
22: SRBase is just like PrimeGrid LLR. Which means Ryzens won't do as well. Worse, their queue is currently filled with itty bitty tasks, so it might not be good for bunkering.

Asteroids@home can get some extra power from Nvidia GPUs. But not much. So is this a good or a bad thing?

There are five remaining projects I'd call "meh":
4: Citizen Science Grid
8: ODLK
16: RakeSearch
19: Rosetta@home
24: NumberFields@home

Are there any of these we don't want?

 

[crashtech]

I don't know what to advocate for, but I can give a brief summary of my current processing inventory that could help the more knowledgeable decide what to do:
156 logical cores: 96 Ivy Bridge or older, 56 Haswell or newer.
5 AMD GPUs: 4 HD 7970s, 1 R9 290
4 Nvidia GPUs: 2 1060 6GB, 1 1070, 1 1070ti

I don't know if this makes me CPU or GPU heavy. I think it makes me inefficient old junk heavy...

 

[ao_ika_red]

I have very limited experience on those projects, but I've done CSG and Rakesearch in the past and both of them didn't tax CPU resources so much (unlike PrimeGrid's LLR and VGTU) so I vote for them.
For GPU, all three projects seem neutral for my gpu, so anything will do, I guess.

 

[iwajabitw]

Seti uses Creditscrew also. We just won 4th behind Gridcoin and Seti.usa, Germany in the last sprint with 6mill, so i would say thats #2 gpu choice. PG being the #1 gpu option as I see it, is Seive bunkerable, or bunker Gen tasks for 5 days, that could be just 1 task.
Looks like pickings are slim this year. Im "Tahiti" chip heavy, 5 total and then two 980's. CPU wise less than 50 cores...and I have not ran a few of the projects mentioned to have a knowledgeable benchmark. I remember rossetta tasks being long maybe asteroids too, so ill leave that up to those with more cores.
I think we have a better TeAm setup than last year. Can we get a front page announcement? I think we have the potential to finish higher than last year over all.

 

[ao_ika_red]

I think we have a better TeAm setup than last year. Can we get a front page announcement? I think we have the potential to finish higher than last year over all.

We need to ask @Ryan Smith and @CommManagerJ for that. This will be more exciting than TAAT vs TH earlier that ended quite abruptly.

 

[petrusbroder]

Here are the positive results as of now. With positive I mean that someone has said it would be reasonable to run those projects:

CSG: 4
Universe: 4
Rosetta: 3
Seti: 3
Rakesearch: 2
ODLK: 1
Amicable: 1
Asteroid: 1
Yafu: 1 (one "?")
NFS: 1 (one "-")

Please continue to post!

 

[Howdy]

As usual I will put my GPUs on whatever comes our way. The only favorite I have is Seti.

I am GPU heavy and CPU light so I will do what I can. Just don't expect any help on the CPU side.

This is where I am also, but I will run CPU projects to the best of their ability (My Cr_app as it will now be known as) I'll just need some working app_configs to use on the CPU side to help them out is all. I can only hope Yafu is NOT selected, I had ran some in the past and had endless loop of no completion until deletion on my part. (CGG?) CSG I am assuming, is also not a particular favorite of mine either......running some now and they are ranging from 3+ hours to 22+ hours completion depending on project.

TLDR; Let me know what to be ran and give me an app_config if there is one and I'll run it.

 

[Markfw]

Any medical research first. I still have cancer.

 

[Ken g6]

Any medical research first.

So, that's a strong vote for Rosetta. It could also be a vote for CSG if they had DNA@Home work, which they don't, so it's probably not a vote for that. Nothing else available this year is medical.

 

[TennesseeTony]

So far I see no opposition to Rosetta, I'd like to think that makes it a definite choice. So now we only need to figure out the remaining projects.

Regarding Rosetta: By default the tasks are set to run for approximately 8 hours. You can change this in your settings on the web page for your account. The way it works is, the task you are processing really doesn't take that long, it just gets processed again and again and again, until the specified time limit is up, then it finishes that final run. The longer the task runs, the more scientific value the result has, as more folding permutations are simulated. But, you can set it to run for as little as one hour, I believe. The points per hour do not seem to vary appreciably, based on whether you let it run for one hour or 72 hours. This makes it a good project for both short duration Sprints, and long duration Marathons.

 

[StefanR5R]

@petrusbroder, I think Ken hinted in #10 that PrimeGrid has 1 vote.

My own preferences are by project topic first, and technical factors second at most:

1. Medicine: Rosetta@home
2. Biology: Citizen Science Grid
3. Astronomy: undecided between SETI@home, Asteroids@home.
   edit: maybe also Universe@home, but only if they improve a lot over their current alpha stage. (Need to check Universe@home whether the mentioned problems are still there.)

The other 8 (eight!) suggested projects all deal in Mathematics/ Number Theory, and therefore do not get my vote. Not because I thought Number Theory wasn't important to mankind's advance (it is), but because perhaps 99 % of my DC output happens in competitions now, and over the year there are already more competitions in mathematical projects than I can shake a stick at.

 

[biodoc]

Rosetta is also my number 1 pick. Their protein engineering work has been outstanding over the years.

I have no preferences for any of the other projects.

 

[petrusbroder]

As it stands now:

CSG: decided
Rosetta: decided
Universe: 4
Seti: 4
Rakesearch: 2
ODLK: 2
Amicable: 1
Asteroid: 2
PrimeGrid: 1
Yafu: 1 (one "?")
NFS: 2 (one "-")

 

[Ken g6]

Well, between issues with servers and my personal preferences, I feel like I've narrowed it down to my four:

1. PrimeGrid, because.
2. Rosetta
3. Asteroids, because Nvidia, and it's amazing how they can find shapes from how a point of light blinks.
4. ODLK, just because I don't like any of the others.

 

[Ken g6]

Say, @petrusbroder, why don't you post a poll for this thread? With the ability to select up to 4 options.

 

[StefanR5R]

To clarify my post #20: I am not voting for Universe, as long as there is no sign of this project to be able to stabilize as quickly and as dramatically as needed to be even eligible for Pentathlon by the organizers' own criteria.

(Editing post #20 now.)